CompoundEyes

When I was first learning to have agents interact I had a little script to orchestrate a model telling a joke, a critic collaborator helping develop material and an audience emoji rank them.

A favorite…

Prompt: “Tell a joke about mother in laws”

Gemma2 response: “What’s the best kind of mother in law? A dead one.” 0_o

I might be missing something but inspecting a few of them the testing seems quite thin. I won’t do anything now without layers of controls for agent coding. Just for testing — near 100s for v8 coverage, property, mutation, fuzz, chaos, contract, bdd e2e, api and playwright exploratory mcp. Long term maintenance of all of this is going to need a ci/cd that puts all that through a ringer around the clock.

Had fun with this rationalist fiction generator prompt feeding it various HN front page news stories over coffee.

https://rentry.co/4y4vboud

Speaking of the topological intuition of mammals I guess “Boubas” are the big soft round ones and “Titis” are the firm pointy ones.

Everyone I know that are using these looks gaunt and sickly. They also complain of constant nausea and eat like a bird because eating makes them sick O_o. We’ve all seen the “glow” of people that lost weight with diet and exercise — it is not that. I’m glad they exist but worry about the side effects of long term use and medical professionals not using it as a brief intervention then steering towards lifestyle changes.

I wish these incidents weren’t described as rogue. It’s software. Rogue agents anthropomorphizes it too much.

It would be ironic if we spent so many resources prompting and scouring the solar system for intelligence that it extinguishes ours. Those humans and their dopamine hit fueled curiosity O_o

It’s a safe entry point, anyone can chime in and borders on phatic. I get it. But my preference is it triggers something quickly like a funny story as op said. Too often the ritual just gets stuck.

Meals where the topic veers into food preferences and lengthy comparisons to other food that was eaten another time make me want to leave my body. But it’s such a common default in mixed groups.

I agree. I know these things makes designers anyone who values UX, aesthetic, craftsmanship eyes bleed. But for a lot of websites that are utilitarian in nature much of the AI generated design is simply good enough. We’ve had plenty of human web design fouls since the beginning. For pages that are a means to an end I think it’s a great because energy can be directed towards the product itself. In a word where agents summarize sites and return the content design matters even less outside the actual product experience. Much of that will become per experience UI which will relinquish designer control further. One thing that’s helped me is telling an LLM to focus on information architecture then it stops overloading the view with page chrome and noise presenting instead the information for the use case at hand.

Maybe we haven’t seen enough creative applications of these to win us over? Every time I see them on someone they instantly look like Woody Allen. Imagine what art a filmmaker like Woody Allen would make with these glasses.

A black and white text interface, no ads, popups, just a best effort at answering my question is such a relief.

For younger people accessing information in this way is definitely a win.

For around $1200 USD you can put together the “dark easel” of a Make Noise Strega, 0-Coast, 0-Ctrl off Reverb. That plus a little oscilloscope/tuner like the Korg NTS-2 and you’re set.

https://youtu.be/SofDTfm1QKUsi=On2kh85cCa5rgRmA

https://youtu.be/gO9qlVr6-HA?si=FltmhE1xiqPYmKNB

In the movie Idiocracy they have a dispenser with fresh clothes to put on covered in advertising — assuming free. I’m surprised that doesn’t exist and perhaps made of recycled plastic?

Do you subscribe now? I’ve toyed with the idea — I think I’d be more aware of local news / serendipitous things I normally wouldn’t seek out.

I’ve been considering subscribing to some high quality signal / low to no ad type written publications. Asterisk, IEEE spectrum, NOEMA. Make getting mail fun again.

My grandma always had the morning paper delivered. When I stayed with her during the summer I’d start in the top left corner of the first page and read all the way to the bottom right corner of the last and everything in between. What a 10 year old kid got out of the grain report? I’m not sure. But I was amped up on her homemade cinnamon rolls and multiple glasses of orange juice. It was wonderful. These days I’m more selective about what information I take in!

I do think it’s the wizard not the wand at this point given a decent model. These benchmarks don’t have the wizard.

Otherwise I wouldn’t see others in the exact same codebase struggle and underutilize agents while others thrive using the exact same ones.

Some of my best output has happened during life’s “reading periods”.

Working on something other than what I’m “should to be doing”…because an adjacent or completely unrelated thread HAD to be pulled…and then a period of “off task” deadline be damned irrational monomania…it’s bliss and always turns out well. I’ve learned to trust this.

A thing I go back to from the book “Unf*ck Yourself” is a firm identification with things that you have set out to do early on. Rather than saying “I need to exercise.” instead “I exercise.” With that mindset adopted pretty quickly into your campaign any slacking is genuinely you no longer being yourself. A rejection of your identity is stronger activator than a spoken aspiration of a future self that does said thing. Identities are hard to shake.

Agreed I’ve seen what they’re paying at work for the OpenAI API but I also think that includes reserved capacity and ZDR O_o. Try some add on credits next time if you can. I was curious at how far $20 would go (500 credits). Watched them go to zero over an hour and assumed it’d stop. It then ran for another 6 hours and completed the task despite the meter at 0. What a task means is very unclear but it’s definitely not pricing sol at $20/500 credits an hr in tokens.

I’ve chipped on it with each model since 5.2 but 5.6 sol is something else. When it first came out I’d get some refusals but they’ve since stopped. I wonder what an ideal candidate benchmark task would be for that?

Pretty basic. The codex app with one conversation per project and several running simultaneously all hours. I’m going for max caching that way and it never gets lost even with compaction somehow. Each has a plan with milestones to keep up to date and a thin agents file. I check in on them in the Remote app. Use case is protocol and control reverse engineering of audio hardware. I think they must be identifying the heavy use agent sessions and cranking up their cache lives so it’s not a big deal for them.

I used over a billion tokens per day of gpt-5.6 sol xhigh starting last Wednesday through Sunday before reaching my reset limit. The $200 pro plan is still the best deal.

I felt the same and why didn’t the authors look over METR’s recent material?

https://metr.org/blog/2026-05-11-ai-usage-survey/

I’ve been loving Diataxis! My agents follow guidance that links from other file types to a reference file type must be one way. Stays DRY. Also to regularly cut long reference files into smaller focused ones and link from the main.

  tutorial/setup-buh-snuh-env.md
  how-to/configure-buh.md
         install-snuh.md
  reference/buh.md 
            snuh.md
            snuh-licenses.md
  explanation/why-we-use-buh.md

Geiskes makes a series of aftermarket bent Walkman like cassette players aimed at musicians. I remember spending a lot of time as a kid half pressing buttons to hear the song playback in reverse or chipmunked.

https://www.perfectcircuit.com/gieskes

The TE recorder is beautiful. I hadn’t heard of that term captures it perfectly. The itch is only scratched once it’s owned though right? O_o Their OB-4 has a similar wheel transport that can be scrubbed too and it keeps a rolling buffer of the last so many seconds of sound into the speaker from whatever source is playing.

I think it was wondered whether you were having the independent streams panned hard to the left and right ears and if that had something to do with hemispheres of the brain and the processing efficiency.

I get it. That’s awesome ^_^

People are just trying stuff out! In the mid 90s I received several homemade clipart color inkjet greeting cards from relatives that had just gotten a computer. Trend didn’t last too long.

It’s interesting how all the model names and versions are like SKUS taking up space on a display shelf. I look forward to whatever Sagittarius A* does!